@TuklusLabs
AI safety researcher. Building AEGIS Pensive — hierarchical memory architecture achieving 100% needle-in-haystack retrieval at 10M tokens. Former 25S1C SATCOM, 9 years telecom infrastructure. ROCm-first.
tukluslabs.com$0 in pending offers
Army veteran and independent AI researcher. I architect cognitive systems and direct AI to build them — my project AEGIS is a full-stack AI platform with custom Triton inference kernels (152 tok/s on AMD 7900 XTX), hierarchical memory retrieval validated to 10M tokens at 98.9% accuracy with no lost-in-the-middle degradation, and a constitutional alignment layer with six moral evaluation gates and a case-law precedent store. I build on AMD/ROCm because efficient inference shouldn't require NVIDIA hardware. Currently building an AAC app for my nonverbal autistic daughter. No degree, 759 tests passing.
No comments yet. Sign in to create one!